Everstack
API ReferenceSandboxComputer Use

Computer Use

Screenshot, mouse, keyboard, and screen recording for GUI automation.

Computer Use

Full desktop automation for vision-capable agents. Requires computerUse: true on CreateSandbox, which launches Xvfb + XFCE4 at boot.

Capability Info

GET /v1/sandbox/{sandboxId}/computer

Returns available operations and whether Computer Use is active.

Screenshot

POST /v1/sandbox/{sandboxId}/computer/screenshot

{ "format": "jpeg", "quality": 85 }

Response: Content-Type: image/jpeg binary body (~50-100KB for JPEG).

Formats: jpeg (default, smallest), png (lossless).

Displays and Windows

GET /v1/sandbox/{sandboxId}/computer/displays
GET /v1/sandbox/{sandboxId}/computer/windows

Mouse

EndpointBodyDescription
POST .../mouse/click{ x, y, button, double? }Click (left, right, middle)
POST .../mouse/move{ x, y }Move cursor
POST .../mouse/scroll{ x, y, direction, amount }Scroll (up/down)
POST .../mouse/drag{ from: {x,y}, to: {x,y} }Click-and-drag

Keyboard

EndpointBodyDescription
POST .../keyboard/type{ text }Type text (handles uppercase, symbols)
POST .../keyboard/key{ key }Key combo e.g. "ctrl+c", "Return", "F5"

Screen Recording

EndpointBody / ResponseDescription
POST .../recording/start{ label, fps? }{ recording_id }Start recording
POST .../recording/stop{ recording_id }Stop and finalize MP4
GET .../recordingsList recordings
GET .../recordings/{id}/downloadDownload MP4
DELETE .../recordings/{id}Delete recording

As agent tool calls

The screenshot operation is available as an MCP/agent tool call:

sandbox_screenshot(sandbox_id, format?)  → base64 image

Image requirements

ToolPurpose
Xvfb, xfce4Virtual display + desktop
xdotoolMouse and keyboard control
scrot or imagemagickScreenshot capture
ffmpegScreen recording
wmctrlWindow enumeration (optional)

Install: apt-get install xvfb xfce4 xdotool scrot ffmpeg

On this page